Bots Before Thoughts | SheFi Summit Bangkok

Transcript

[Music] okay so let's start with intros because this panel is stacked and I want to hear all about you all so hey everybody go shivi my name is Sydney Li and I am one of the co-founders as well as head of developer relations at GAO hey uh I'm Liz I'm a partner at a6z crypto hi I'm Maria I'm head of marketing at life pure and I'm Tina I am product Leed at hyperbolic flaps which is an inference and GPU Marketplace so I'm super excited about this panel because I think there's a very broad spectrum consumer facing

applications and both excitement about Ai and crypto and also institutional excitement and a lot of people specifically I would say my sister and family don't understand any of it and what we want to kind of break down in this panel is a lot of that context and kind of discuss why people are excited about it so kind of the first question I'll go into is what do you foresee is kind of the future of AI and crypto and why you're both excited in about it in your role and then um for the future of maybe say consumer adoption or the

ecosystem yeah I can I kind of get started there I think uh one of the things I'm most excited about in this intersection is the idea of agents and why and how agents can interact on chain I think a lot of the things that make crypto really hard for consumers make it really great for computers and for Bots and agents to use um and so I'm pretty excited about as we see more and more agents populate the internet using crypto as a means to kind of orchestrate how those agents interact um what are the rules of the road what do you do when two agents run into each other I think you know code is law as a complicated concept for humans but a very easy concept for agents and for what Bots and so I'm excited to see that kind of play out and see how we can use

crypto as an orchestration and also a way to keep control of and create rules for agents as they run free on the internet that's a really big topic I think to provide context we first look at why AI even began in the first place and so the working thesis about why are we pursuing AI as humanity is that we're trying to get towards AGI or artificial general intelligence the concept that AI can perform like humans but better smarter faster and I think one of the differentiators that you talked about is in why web 3 and Ai and the concept there is that to what you were saying Liz earlier is that we want to automate self-run tasks so most of web 3 wants to understand how do we use specific tasks to be orchestrated and at this current

rate we are doing so in using smart contracts which is like Hey how do we have an atic framework and then how do we settle those specific decisions when certain conditions are met and I think that it's going to probably require something that is more of a hybrid based approach which would not be more of a definitive subtle I'm happy to talk more about that as well I would Plus on all of that stuff um and then I would also add a piece of governance I think AI struggled with um this black box of decision-making I think blockchain could really help with that um and then on the other hand AI has do that mass adoption Al a negative brand in some cases um I

think the two together could bring M Mass adoption to blockchain as well it would be amazing I love just hearing from Smart women because breath of knowledge and I love syney you mentioned automation workflows I think Liz you mentioned trustless execution environments and then you also re mentioned um developer adoption and governance automation could you guys um take an existing onchain workflow or offchain workflow or a use case that you can think of that AI specifically enhances or revolutionizes because I think one thing a lot of people in existing web 3 spaces kind of struggle with is visualizing the actual um

improvements and what they'll be very heated about this topic I know I'm like my heart is racing I've got thank you for laughing because it's real so when we're talking about how do we actually see this in a real world use case y'all a hackathon we had a hackathon recently and the manual process of slacking the colleague like hey sorry can you put can you put 20K in the no is safe no you didn't respond okay let me just send you a calendar invite you didn't come to the calendar invite okay so then I have to go to telegram like sorry guys the the money's coming I'm so sorry right so this is still extremely manual which is like just simply paying out bounties for a haathon and so then it's like how do we actually automate these self-run

tasks so again as I was saying earlier I think AI is interested in AGI I think at least here in web 3 at this current state we are interested in automating certain self-run tasks and the last thing I'll say and this is because you talked about this earlier about governance is that right now governance is still pretty manual right still pretty manual and what that means is that governance data sets are still socially swayed so what we've done for example is we have data sets across like let's say 350 Dows we can see these delegates of so and so L2 Dow they always vote the same way these last three delegates they're always voting based off consensus they don't have their own opinion right so then the question is how do you actually create

automated orchestration so that you can make decisions Beyond just manual like current manual programming which is how logic is currently being created yeah I think I think talking about these real use cases you kind of have to break it down into two different sectors so you have how can AI help crypto and you have how can crypto help Ai and so in the AI helping crypto world I think this is a little bit of what you're talking about right now but also simple things as easy as how can we help developer adoption how can we use tools like cursor but to help with specifically smart contract auditing or other kind of like web 3 specific use cases um how can we lower friction how can we make the consumer interfaces better how can we make ux

better that kind of all falls into the how can we use AI to help crypto I think on the other side of how can we tangibly use crypto to help AI there's a number of really interesting kind of spaces that are being explored I think looking at things like authenticity provenance um how can we kind of track down fake news how can you track on fake images that's like one big area that's that's being explored um how can you use blockchain very simply as a transparent and immutable ledger to kind of you do that provenance I think the other one that's really interesting is going back to what I was talking about agents and payments um and so I think you know we saw a very good example of this on Twitter with you know truth terminal spinning up a wallet and being able to get invested in and

then launching meme coins and all that craziness and I think a lot of that is you know fun toy hype world but I think it is indicative of a larger example of if we're going to have agents running free on the internet performing these automated tasks they're going to need to have wallets they're going to able to be able to pay for things they're going to be able to transact um and I think that is a very tangible thing that's happening right now where crypto is really helping AI I I used to work at a traditional bank trafi and we always talked about what was the most largest unbanked population agents will be the most UN like unbanked population coming up yeah I was just going to say um I 100% agree with that I think you'll hear that a lot 100% agree with everything

you know I think we need to reframe reframe it as Liz is saying it's not just like what AI can do to help us it's like what can we do to solve real world problems and I think there is a lot of scope and Healthcare cyber security you know at life P not to shill but um essentially we are onboarding video AI video developers right and that is a big Market it's already a proven use case so we have to meet the developers where they are and meet people where they are um without shoving that like blockchain narrative I have so much more to say no go is that possible this is good we're in for a ride also like just to Define agents because I think that's a buzzword that not a lot of people know that can be anything from a Twitter bot running

autonomously to an onchain bot running autonomously to a Twitter API that you configure to run autonomously so any that is running autonomously yeah maybe to add to that I just kind of the way I think about it is it is a piece of code that executes an action on your behalf and it can be as simple as like a limit order it can be as complicated as a twitterbot that's tweeting for itself and spin up wallets and anything in between that but generally when code is autonomously ex exting actions on your behalf in web three terms it is a Dap with AI capabilities yeah I think you talked about you know the proven market for Life Pier video I think another thing to understand is to like that Chasm between

AGI and currently self-run tasks for these AI agents like we're really harping on this right and and why because at this current rate of how AI is being developed the core fundamental is that it still relies on multimodal integration so this multimodal integration that means is the ability to contextualize and generate across text images video audio but also be able to generate and process the cultural context and Nuance Behind These images and these videos and then so what that means then is then having to figure out a way to be able to handle abstract logic and Abstract planning and so we're not quite there yet and I think where

we're starting right now is with these agentic self-run task completion okay I love that you brought up AGI because that goes into my next question but a lot of the traditional AI world right now is kind of going into this deeper discussion about ethics and um the ethics involved with aggi and that same conversation as a result of Automation and AI coming into crypto is also arising now as well and I want to know what you guys specifically think about the ethics of building this infrastructure in crypto and what that impact would potentially be yeah I think I think it's a good question um I think one of the great things about web 3 is

that it's incredibly deterministic and so I really like the idea that you can use web 3 or blockchains to put guard rails on AI in the way that AI is incredibly non-deterministic for the most part um and you can kind of let it run free and that's what people are afraid of you can do the exact opposite in crypto and so I I like this idea of putting guard rails on what an AI may or may not be able to do using cryptography or a blockchain but more generally I think that building with web3 ethos is the safest way to build AI having everything be open sourced having it be transparent having everyone be able to have a say in it not being Behind These walled Gardens not being hidden um I think we haven't seen you know a ton of in as or as much interaction as we would

like with something like a dow and voting but something that people clearly care a lot about is how are we censoring or not censoring these AIS how is you know gemini or any of these models being drained or being have Prejudice put in or taken out or bias and I think that's an area where it might make a lot of sense to have a doubt and let people vote on what should or shouldn't be allowed and what should or shouldn't be said and so I think that whether or not we're actually literally building on chain with AGI building in and around that ethos I think is probably the safest way to go about it um but I know a lot of people disagree so I I was actually just going to say I personally would disagree only because well one there's various schools of thought to

getting to AGI there's various schools of thought and I think right now the the terministic thought you were talking about is in web 3 we're seeing leading zkl solutions to creating this type of governance settlements as we already know everyone knows that smart contracts are great because it settles everything definitively right I I personally am not sure if that kind of what I said earlier is enough to be able to contextualize the multimodal integration which is you need to be able to contextualize images text videos Etc and so this this actually keeps me up and I I recently had a conversation with vitalic to double check the working thesis that our team was having which is how do we build

something maybe more of a hybrid contract model I'm not exactly sure what that would look like right and you know we're willing to spend the time to find that out and uh yeah I think that you know from from his perspective that AGI at one point and this is kind of like Beyond web 3 we'll kind of bypass all that we're trying to do right now which is the Smart contract solution but uh yeah yeah yeah know I I totally agree that it's not a given or even I think particularly necessary to literally put the AGI on chain I think it's more how can we use these kind of as a method of orchestration between whatever is running on your computer whatever is running on my computer how I want to interact with the internet how you want to interact with the internet more so

than literally an entire model put on ethereum I have nothing to add here I was going to say I agree with Liz but then you explained it I'm like say more say tell me more okay so one thing Sydney went into it's a library called CK ml it's an incredible Library I think coined by some of the world coin contributors and there's a number of other web 3 contributors but it primarily does research on um onchain verification and onchain model fine-tuning as well as onchain use of models in ethereum and right now the confirmation time for using ckl and opml the two um ml or machine based machine

learning based Frameworks in ethereum typically are 11 to 15 minutes last time I checked might be wrong but how do you guys feel about the fact that a lot of the onchain verification and model development is not at the place yet to build a lot of this long-term infrastructure and what you foresee with the current development and how it affects it I'll just I'll just jump in there's several things to tackle I ladies ADHD I don't even know where to be in um okay so we could talk about the zkl we could talk about scaling right those are kind of like almost two different questions so the zkl side is

essentially the concept of using machine learning and then adding a security layer again if we're using like a comparison Apples to Apples kind of we can look at it how AWS is tackling it and so from AWS how they tackle ZK is the example of let's say um automated reasoning so this type of reasoning is like uh a cat has hair fluffy is a cat and then therefore fluffy has hair or fur excuse me I'm sorry fur right so like that that that type of that type of reasoning and then adding that type of reasoning and securitizing of of an of ml model for example and then now we're talking about okay that's one piece and then the last thing I'll about that is that it's still very deterministic what we talked about earlier which is it's

it's kind of hard to have something deterministic but then does cat actually have fur or is it one of those bald cats right so I don't we don't know right so there's that one piece and then the second piece is then the the scale the time that we were talking about and so that's something that we're currently addressing right now so in the last three weeks we have been working with agen layer specifically here in Thailand to figure out how do we scale this and how do we secure this right and then we also found out that during this time as we're building with mega eth like hey can Mega eth as its own I don't want to call it L2 but like how can you actually settle that even faster and bring it down for right so I'm not here just to name drop I'm just here to give an

example of what we're tangibly looking at to kind of address this uh 15-minute Gap can I ask a follow-up question for the audience so when you reference Mega eth and aan L are you trying to collaborate with them because they have a larger sequence ER so the kilobyte and transaction size is larger for these model requests that's a really good question I think it's for us because for us specifically we are our we are so guia is a developer platform where it's open source and you can build apps on top of data sets and we're the closest analogy is to Docker we're three uh four lines of code and we we are able to download and containerize your whole environment so you don't have to like

peace meal everything it's like know instead of grocery shopping for dinner you just buy a meal kit and then it's like oh it's here and then you just build a meal but I think the question with agan layer specifically and then Mega eat is um similar to our like mental framework it's about being able to compartmentalize and package that so much more efficiently and effectively uh yeah yeah I think maybe going back a little bit to your original question I think it depends a lot on why we're using zml right like what are we trying to get out of this are we trying to run on chain do we need verifiable compute are we trying to you know like there's a number of different reasons that we might or might not want to use some sort of cryptography to verify some output

from a model or some distilling of a model or training of a model or there's like a number of different kind of pieces that you might want to verify and all the way from a simple inference request and a tiny like XG boost or like a full training run of an llm like of the order of 10 billion parameters right and so depending on where in that Spectrum you are and then how High your security is or how high your need for security is we have a wide variety of kind of like options on the menu and their ability to scale and so if you're talking about like I want you know consensus level security to train you know gp6 I think we're very very far away from that um if you're talking about I want some sort of like

realistically optimistic security for inference because I'm using a decentralized compute platform because it's cheaper than going to AWS right now I think you can probably lower your security threshold a little bit and so it might be more feasible um or it is more feasible today and so I think that naturally we'll see the things scale but a question that I bring up a lot is like what what is our goal here right like what is the goal with putting this on chain or what is the goal with getting verified blence here and how does that change the approach how does it change how realistic it is I think we're a very long way off like I said from seeing an chatu te like product on Chan but I don't think we're that far off from seeing a much more kind of reasonable use case for AI I would I would agree

with that it's like from a tactical perspective there's like use cases that we can just go into but if we are going longer term longer Vision I don't know years okay and my next question is more so about adoption so I think one thing crypto and web 3 has struggled with is long-term sustainable adoption there's been DSL languages that are distributed systems languages that are built on crypto photography that are trying to get developer adoption there's been all VMS that are also building out languages or ecosystems and trying to get adoption how do you foresee the both AI infrastructure and the AI research

that's build building out right now in our community being sustainably uh maintained and how do you see developer adoption um kind of maintained longterm so the question is like how do we bring more people yeah with competing ecosystems and products that have also struggled to have developer adoption how do you think that a crypto and AI can specifically have that interest or sustain it do you want to go first from a de perspective oh yeah oh yeah head of dead that's right developer adoption so I I think without a doubt the ease of AI adoption is the ux it's a chat you just type in in the chat right and then of

course we look at web 3 it's a picture you just paid $20,000 for it so when we look at the different ux experience may it be an nft or if it's a chatbot it does require several several several um levels of abstraction right a lots of lots of abstraction it's just a chat bot it's just a JPEG pay for it now then we look at developer adoption I think that developer adoption I'm guilty of this I'm sure those who you're like I have a problem problem I'm going to make a solution and you just over engineer it you just over engineer it it's like you're trying to make the fanciest dinner and like girl just put noodles and some water like go to bed right so we tend to over engineer things or we

don't want to uh we don't want to we like as developers we're like oh we're we're too smart we can we can figure this out so I think that when it comes to how do you break the chasm of adoption is you really need to understand who do you need to adopt and as of right now I think web 3 AI is limited to what's validated in the current economic market and that is let's say delivering a product for developers and so therefore that means that teams might have one or less ux designers on their team right so I think that we are a in this industry currently a developer or information Le first go to market versus like a product you actually go to market yeah I would say

um okay I'll go first I think it's about creating value before you asking for value and meeting the developers where they are um tons of education is required for both Ai and blockchain and then beyond that it's um you know I've seen this happen in gamey narrative too right like I said this earlier we're saying like blockchain blockchain blockchain and it turns people off I think we need to move away from that and like this is the problem that we're solving this is just the underlying technology to solve it yeah no I I agree I think what I what I was going to say is finding a developer that is both like an expert solidity engineer and also can build models is is maybe not impossible but very very tricky and so building a product that ask asks your developers to know both is going to really struggle is

going to be a very hard time for you and so understanding whether your main kind of like Target developer is a crypto engineer and you're going to abstract way the AI is an AI engineer and you're going to abstract with the crypto or maybe it's not a developer and then we're this a different conversation um but I think where companies get into trouble is when they're trying to Target developers that both need to deeply understand Ai and crypto because the V diagram is just so small and those people are probably already getting paid a lot of money to work at a different company um and so one at least one needs to be abstracted away you cannot be everything to everyone with free marketing put that someone please tweet that yeah I was just going to say I think that this this specific this specific example really hits home

because I think as developer framework again four lines of code so we're not we basically abstract the need for you as developer to know to need to know either web 3 or AI you're just a developer and you need to build some sort of decentralized AI so that's something that's close to us okay for last question I will just give maybe you all touched on maybe safety aspects of AI and I think Sydney touched on coinbase STK and one of you also mentioned verification of personhood I mostly just wanted to wrap it up and kind of go into what you guys think about guard TRS and safety of these tools and infrastructure

that they were building and how both the women in here that would potentially be impacted by it can be mindful of that yeah I mean I think that what you just brought up is a really great example of what we were talking about earlier of how you know crypto can be the guard RS in AI so we've talked a lot about you know agents and how great they are going to be and you're all going to have these agents deployed for you doing a million tasks that you don't want to do and it's going to be amazing but also you know that turns into an Internet run by Bots um and so I think that what crypto can do to help is this verification of personhood that you were referencing and so you know if we can have a kind of differentiation such that humans can

remain first class citizens on the internet even while you have an army of bots doing all the things that we want our army of bots to do that we don't want to bother with um that's going to create a huge kind of differentiation allow for safety on the internet while still allowing AI to do what AI is good for um and I think that what we've seen is really the only way to do that is through crypto um to have really strong verification of personhood and onetoone my am myself I you know hold an nft that is me um versus a bot and even if I want to delegate that person to a bot at most I can do is one to one yeah and and also that's the huge differentiator is that if you assign an ID to an AI agent that differentiates the agent from a

bot plus one okay thank you so much guys everyone give a round of applause for the lovely panelists amazing